On 9/16/2016 1:04 PM, Jason Gunthorpe wrote:
> On Fri, Sep 16, 2016 at 12:56:55PM -0400, Hal Rosenstock wrote:
>> On 9/13/2016 2:03 PM, Jason Gunthorpe wrote:
>>> opensm is not qualified since it rides on the other libraries.
>>
>> I'm not sure whether opensm should be in or out. However, I don't
>> understand the not qualified part. It does seem to be similar to both
>> srptools and ibacm which are all daemons and live on top of libibverbs
>> and libibumad (ibacm has addition netlink socket support).
>
> srp_daemon and iwpmd are both bits of code that reasonably could have
> lived in the kernel but have been punted to user space to keep the
> kernel simpler. For instance srp_daemon's job is to collect service
> record from the SM and tell the kernel about it.
>
> ibacm can reasonably considered a linked subcomponent of librdmacm.
>
>> What are the metrics for inclusion ?
>
> Unclear :)
>
>> What other libraries are you referring to ? OpenSM lives on top of
>> libibumad and two internal libraries (complib and a "vendor" library).
>
> Those are the libraries.
>
> opensm is not bound to kernel, does not use the kernel 'uabi' headers,
> is not a part of the kernel that was delgated to userspace, and is not
> required/related to any of the subcomponents already in the tree.
Not to mention it's also a complex beast ;-)
--
Doug Ledford <dledford@xxxxxxxxxx>
GPG Key ID: 0E572FDD
Attachment:
signature.asc
Description: OpenPGP digital signature
